HiSeg: Unfolding of Segment Hierarchies in Color Images

Image segmentation is one of the important components in many image-processing applications. Despite many researches on image segmentation, it still remains as an unsolved problem. This is mainly because identifying objects from an image data is hard task to do. In this paper, a new segmentation scheme is presented which segments the input color images in two soft and hard phases. After preprocessing the input image, a soft segmentation method is applied in order to segment the input image into initial small segments. Then, a hard segmentation phase starts by constructing a weighted network from the soft-segmented image and the communities of the network are extracted. Each resultant community in the hard segmentation phase represents a segment in the input image. Finally, a post-processing phase is done on the result of the hard segmentation phase. Parameter freeness is a very nice property that gives significance to HiSeg. To demonstrate the real pure result of HiSeg, a comprehensive sensitivity analysis is done on it. In addition, the results of HiSeg are demonstrated, and compared with some existing segmentation algorithms qualitatively and quantitatively. Extensive experiments have been performed and the results show that HiSeg can reliably segment the input color image into good subjective criteria.